you can let them evaluate preproduced LLM output to get an answer to that question.